CN109550712B - Chemical fiber filament tail fiber appearance defect detection system and method - Google Patents
Chemical fiber filament tail fiber appearance defect detection system and method Download PDFInfo
- Publication number
- CN109550712B CN109550712B CN201811635403.8A CN201811635403A CN109550712B CN 109550712 B CN109550712 B CN 109550712B CN 201811635403 A CN201811635403 A CN 201811635403A CN 109550712 B CN109550712 B CN 109550712B
- Authority
- CN
- China
- Prior art keywords
- image
- silk
- paper tube
- label
- ingot
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
- B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
- B07C5/3422—Sorting according to other particular properties according to optical properties, e.g. colour using video scanning devices, e.g. TV-cameras
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/89—Investigating the presence of flaws or contamination in moving material, e.g. running paper or textiles
- G01N21/892—Investigating the presence of flaws or contamination in moving material, e.g. running paper or textiles characterised by the flaw, defect or object feature examined
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Textile Engineering (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a chemical fiber filament tail appearance defect detection system and method. The system comprises a tray for loading the silk ingots and a conveyor belt for conveying the tray, wherein the silk ingots are provided with labels, a camera bellows is arranged on the conveyor belt, a sorting unit is arranged on the conveyor belt behind the camera bellows, an image acquisition unit for acquiring label images and silk ingot images is arranged in the camera bellows, and the image acquisition unit sends acquisition information to a processing unit for tail silk defect detection. The system identifies and reads label information from the label graph, screens an image to be detected with a silk spindle paper tube from the silk spindle image, carries out edge detection on the paper tube area of the image to be detected, extracts a main characteristic description paper tube area, brings the processed image to be detected into a tail fiber classifier for classification, obtains silk spindle tail fiber detection information and sends the information to a sorting unit, and the sorting unit sorts good and defective silk spindles according to the classification of the silk spindle tail fibers. The invention saves cost, has higher working efficiency and more accurate detection.
Description
Technical Field
The invention relates to the technical field of spindle defect detection, in particular to a chemical fiber filament tail fiber appearance defect detection system and method.
Background
Chemical fiber filament fiber-free is mainly produced by four factors. The operation reason is as follows: the head is raised, the tail is left, and the head is not well left, possibly because of the tendency of people. The process factors are as follows: the process set up for the retained fibers is not appropriate. Paper tube factors: the length and depth of the tail fiber groove and the angle of the notch are not proper, and the length of the paper tube is also not proper. Equipment factors: some domestic devices do not retain the fiber easily.
At present, the production lines of most chemical fiber spindle production plants adopt manual detection modes for appearance observation, but the manual detection is high in labor intensity and low in production efficiency, the manual detection can only be carried out after a fiber tube is taken off the machine, the subjective performance of people can directly influence the defect detection quality of products, and the data of manual detection cannot be accurately and timely brought into a quality management system to form production quality assessment for the whole product batch, so that the traditional manual detection mode has hysteresis, the downstream processing performance is influenced, and the real reason of the defects cannot be accurately and timely found to eliminate production and management faults. The production of chemical fiber ingots is a production process with high speed and high automation, the traditional artificial defect detection can not meet the requirement of fine production, and the online tail fiber detection system based on machine vision and image processing technology can effectively ensure the detection precision of defects, generate a product quality statistical evaluation report in real time and assist in standardizing production and management processes.
At present, the varieties and grades of silk threads in the chemical fiber industry are various, industrial robots replace production workers to finish heavy work, but the appearance detection is not greatly broken through, a patent specification with the application number of 201210049619.2 and the name of a silk bundle appearance quality detection system discloses a silk bundle appearance detection system. Patent specification CN201510883141.7 entitled "surface inspection apparatus and method" discloses an appearance inspection method based on image processing, which reduces waste wire interference by an air blowing unit, but there still exist some waste wire interference situations that cannot be removed by the air blowing unit, and the invention only ensures inspection and does not solve the problem of classification in the later period.
Disclosure of Invention
The invention mainly solves the problems of high labor intensity and low production efficiency of manual detection in the prior art and the problems of low automation degree and inaccurate detection of a general detection system, and provides a chemical fiber tail fiber appearance defect detection system and method.
The technical problem of the invention is mainly solved by the following technical scheme: a chemical fiber tail yarn appearance defect detection system comprises a tray for loading a yarn spindle, a conveyor belt for conveying the tray, a label arranged on the yarn spindle, a dark box arranged on the conveyor belt, a sorting unit arranged on the conveyor belt behind the dark box, an image acquisition unit for acquiring a label image and a yarn spindle image arranged in the dark box, and a processing unit for detecting tail yarn defects by sending acquisition information to the image acquisition unit;
the processing unit is used for identifying and reading label information from the label graph, screening an image to be detected with a silk ingot paper tube from the silk ingot image, carrying out edge detection on a paper tube area of the image to be detected, extracting a main characteristic description paper tube area, and bringing the processed image to be detected into a tail fiber classifier for classification;
and the sorting unit is used for sorting the good and defective products of the silk ingots according to the classification of the tail fibers of the silk ingots.
The invention adopts machine vision and image processing technology to detect the tail filament of the filament ingot in the process of conveying the filament ingot, and arranges the sorting unit to remove the defective filament ingot with the tail filament, the whole detection system does not influence the normal production of the filament ingot, and a complete set of coherent on-line processing flow from image acquisition to defective filament ingot removal is realized. According to the invention, the image acquisition unit and the processing unit extract the label information through an image processing technology, so that the installation of redundant hardware equipment is avoided, the cost is saved, the identity of the silk ingot is determined by sequentially photographing, and the defect detection of the silk ingot is completed. The invention utilizes the motion rule of the transmission belt and realizes timely, accurate and high-speed elimination of the defective silk ingots by installing the sorting unit.
The system establishes a tail fiber classifier before detection, and the establishment of the tail fiber classifier comprises the steps of collecting a large number of silk ingot images by a processing unit, respectively carrying out paper tube region detection and paper tube judgment on the silk ingot images to obtain the silk ingot images with paper tubes, then carrying out edge detection on the paper tube regions of the silk ingot images, extracting main feature description paper tube regions highlighting tail fiber features, and then substituting the silk ingot images with the main features into a bp neural network for training to obtain the tail fiber classifier.
The image acquisition unit includes that the first unit of making a video recording that gathers and the second unit of making a video recording that gathers each side image of silk spindle are gathered to silk spindle label image, and first unit of making a video recording can be a camera, and the second unit of making a video recording includes camera group and light source group, and camera group includes a top camera, two side cameras and two bottom cameras. The cameras are CCD cameras, each camera is internally provided with a photoelectric sensor capable of acquiring high-precision images, and the high-precision images of moving objects in a stable time period are acquired by combining an external trigger scanning mode and controllable exposure time. The light source group is mainly used for stably polishing the filament ingot, can be arranged at a proper viewing position of the camera group, and preferably comprises a top center vertical light source, a bottom left annular light source and a bottom right annular light source, wherein the three light sources are respectively fixed by a four-angle type angle locking device to prevent the light sources from shaking, so that the image imaging is unstable. When the filament ingot reaches the exposure range of the camera, the light reflected by the filament ingot is projected onto the photoelectric sensor through the camera lens, after the photoelectric sensor is exposed, the photoelectric diode is excited by light to release electric charge to generate an electric signal, the electric signal controls the current generated by the photoelectric diode by utilizing a control signal circuit in the photosensitive element through a camera chip and is output by a current transmission circuit, the camera chip collects the electric signal generated by primary imaging and uniformly outputs the electric signal to an amplifier, the electric signal after amplification and filtering is sent to an analog/digital (A/D) circuit, the A/D circuit converts the electric signal into a digital signal, and outputting the images to a Digital Signal Processor (DSP) of the camera, carrying out post-image processing such as color correction and white balance processing on the images by the DSP, encoding the images into image files with specific resolution and image format supported by the DC, and finally saving the image files to a memory.
And a tray fastening mechanism for fixing and rotating the silk ingots is arranged on the transmission belt and is controlled and connected to the processing unit.
As a preferred scheme, the rear end of the conveying belt is forked to form a good product conveying channel and a defective product conveying channel, the sorting unit is arranged at the forked position of the conveying belt and comprises a base plate, a first stop lever and a second stop lever are arranged on the base plate and located in front of the good product conveying channel and the defective product conveying channel respectively, the first stop lever and the second stop lever are connected to an air cylinder respectively, and the air cylinder is controlled to be connected to the processing unit. The sorting unit can be controlled by the processing unit to operate, when the filament ingots are classified to have tail filaments, the filament ingots are judged to be good products, the first gear rod is controlled to descend, the filament ingots are conveyed to a good product conveying channel, when the filament ingots are classified to have no tail filaments, the filament ingots are judged to be defective products, the second gear rod is controlled to descend, and the filament ingots are conveyed to a defective product conveying channel.
A method for detecting appearance defects of chemical fiber tail fibers comprises the following steps,
establishing a fiber classifier for detecting the presence of a fiber, comprising the steps of:
s1, collecting a large number of silk ingot images through an image collecting unit;
s2, carrying out paper tube area positioning and segmentation on the silk ingot image; and preprocessing the image of the silk ingot by positioning and segmenting, wherein the preprocessing is to use median filtering to denoise the image of the silk ingot.
S3, classifying paper tubes in the paper tube area; and judging whether paper tubes exist in the paper tube area or not by using a trained two-classifier.
S4, carrying out edge detection on the paper tube region image classified into the paper tube, and extracting the main characteristics of the tail fibers;
s5, carrying out tail yarn classification on the silk ingot image of the paper tube area processed in the step S4, and sending the well classified silk ingot image of the paper tube area into a bp neural network training tail yarn classifier; the tail fiber classifier judges the input silk spindle image, judges whether tail fibers exist or not, and completes detection of the chemical fiber silk spindle tail fibers. Training classifiers by bp neural networks is prior art and reference may be made to y.le Cun, b.boser, j.s.denker, d.henderson, r.e.howard, w.hubbard, l.d.jack, et al.hand-written digital recording with a back-amplification processing systems, in Advances in neural information processing systems, 1990.
And acquiring a label image of the to-be-detected filament ingot, identifying and reading label information, acquiring a filament ingot image of the to-be-detected filament ingot, processing the filament ingot image in steps S2-S4, and classifying the processed filament ingot image through a tail fiber classifier. The to-be-detected filament ingot image is a filament ingot image for obtaining the to-be-detected filament ingot.
The invention adopts machine vision and image processing technology to detect the tail filament of the filament ingot in the process of conveying the filament ingot, and arranges the sorting unit to remove the defective filament ingot with the tail filament, the whole detection system does not influence the normal production of the filament ingot, and a complete set of coherent on-line processing flow from image acquisition to defective filament ingot removal is realized.
The processing unit writes classification information and daily statistical data on a production line into the electronic tag together after tail silk classification is carried out on the silk ingots, the electronic tag writes the time of the current silk ingots reaching the sorting unit into the electronic tag according to the conveying speed of the conveying belt and the distance from the current position of the silk ingots to the sorting unit, and the time is also written into the electronic tag.
As a preferable scheme, the specific process of positioning and dividing the paper tube region of the ingot image in step S2 includes:
s21, collecting a large number of silk ingot images, labeling a paper tube area by using a BBox-Label-Tool, generating an external rectangular frame of the paper tube area, taking the labeled silk ingot images as a training set, and converting the training set into data in an LMDB format; this step prepares the data.
S22, using the pre-trained VGG-16 network provided in the SSD demo as a basic network of the SSD detection network, and carrying in the data set obtained in the step S21 for training; this step uses SSD training to detect paper tube areas. The VGG-16 network is a deep learning network, and after sample training, the silk ingot image of the standard paper tube area is used as a sample, so that the paper tube area of the silk ingot image can be positioned. The VGG-16 deep learning network is a mature technology, and the specific content thereof can be referred to Simnyan, K., Zisserman, A.: Very deep connected networks for large-scaleimage registration. in: NIPS. (2015).
And S23, detecting a paper tube area of the silk ingot image by using the SSD, and externally cutting the paper tube area according to the detected rectangle.
As a preferable scheme, the specific process of classifying the paper tubes in the paper tube region in step S3 includes:
s31, collecting a large number of silk ingot images, and taking the paper tube area images of the silk ingot images obtained in the step S2 as a training set;
s32, classifying the paper tube region images into paper tube region images of paper tubes and paper tube region images of non-paper tubes, and training the classified paper tube region images by using a VGG-16 network to obtain a paper tube classifier; the paper tube classifier is a two-classifier, the class 0 represents a paper tube, the class 1 represents a non-paper tube, in the detection process, a loss function layer is removed, an output result of the last layer of the network is extracted, and then a sigmod function is used for normalizing the detection score to be 0-1 to serve as a final detection score.
And S33, sending the paper tube region image to be detected into a paper tube classifier, if the paper tube region image is judged to be a paper tube, entering the step S4, and if the paper tube region image is judged to be a non-paper tube, returning to the step S2, and operating the next silk ingot image. All images of the non-paper tubes are removed in the step.
As a preferable scheme, in step S4, the specific process of extracting the main features of the fibers by performing edge detection on the paper tube region image classified as the paper tube includes:
s41, Gaussian filtering is carried out; gaussian filtering smoothes the image and eliminates noise.
S42, calculating the rejection intensity and direction of each pixel point in the paper tube area image;
s43, applying non-maximum value suppression to eliminate the stray corresponding to the edge detection;
s44, determining real and potential edges by applying double-threshold detection;
s45, canny edge detection is completed by inhibiting the encouraged weak edges; in the step, canny edge detection is adopted to highlight the fiber characteristics.
S46, performing principal component analysis on all paper tube area images, wherein the principal component analysis comprises calculating a covariance matrix of features, solving characteristic values and characteristic vectors of the covariance matrix, judging the importance degree of the features according to the size of the characteristic values, selecting a certain number of characteristic values and characteristic vectors as expressions of the images, reducing extracted characteristic dimensions, selecting a certain number of characteristic values and characteristic vectors as more obvious characteristic values and characteristic vectors, namely selecting a set threshold, and judging the characteristic values and the characteristic vectors as obvious when the values of the characteristic values and the characteristic vectors are more than the threshold; the calculation formula of the assistant is as follows:
C＝∑(x-u)(x-u)T
where C is the covariance matrix, x is the pixel value of the image, and u is the mean of the image.
This step describes the paper tube area image, i.e., the sample, with a small number of features that highlight the fibers. The main characteristics are extracted, and the classification speed can be accelerated.
As an optimal scheme, the specific process of acquiring a label image of a to-be-detected silk spindle, and identifying and reading label information includes:
A1. setting each rotation angle of the silk ingot through calculation, and acquiring a label image by an image acquisition unit after the silk ingot rotates by one angle;
A2. pre-building segmentation models
Collecting a large number of label images as a training set, labeling the shape of a label area and an external rectangular frame on the label images, and carrying out segmentation training by taking data into the FCN; FCN (full volumetric Networks for semantic segmentation) is a network structure based on full convolution to realize target segmentation, and adopts an end-to-end segmentation structure, inputs a whole graph and outputs a segmentation result, and the segmentation result expresses a target area in the form of an area set. The technique can be referred to as: long, e.shelham, and t.darrell.full volumetric network for the detailed segmentation. in CVPR, 2015.
A3. Sending the label image to be tested into an FCN test, wherein the test result of the FCN is the position of the label, and obtaining a segmented label area image; and finding out the structure which best accords with the practical application scene as the position of the label according to the ROC curve. The receiver operating characteristic curve is abbreviated as ROC curve.
A4. Pre-building recognition models
Collecting a large number of label images to perform step A3 to obtain a label area image, extracting the characteristics of the label area image by using a VGG-16 network, copying N parts of the processing result of the label area image, and sending the copied result into a recurrent neural network for recognizing the symbols with indefinite length to train; the prior art adopts the cyclic Neural Network for training and then character recognition, and specific contents can be referred to (Long-Short Term Memory (LSTM-RNN), Current Neural Network (RNN), Prediction of Single Stock Price, Intelligent science JOURNAL NAME: JOURNAL of chemical Finace, Vol.8No.1, January 31,2018.
A5. And sending the image of the area of the label to be detected into a recurrent neural network for label character recognition, and acquiring label information.
As a preferable scheme, the step of obtaining the divided label region image and then performing correction in step a3 includes:
A31. cutting a label area image from an original label image according to the FCN segmentation result to be used as an interested area, binarizing the interested area image by using an edge function, and detecting the boundary position of the label area by using the binarized image;
A32. using a Radon corrected inclined character label, calculating Radon transformation of the edge image, and performing Radon transformation on each pixel of the position of the inclined character in the label area image, wherein the pixel is 1; the Radon algorithm is an algorithm for determining an image inclination angle by finding an angle at the maximum projection value through directional projection superposition. The projections are respectively made in the directions of 0-179 degrees. The training is facilitated by correcting the size of the image of the same label area.
A33. Detecting a peak value of g (s, theta) in the Radon transformation matrix, correcting the label area according to the peak value, wherein the relation between a Radon transformation plane and an element plane is as follows:
where f (x, y) is the gray scale value of the point (x, y) on the original image plane, g (s, θ) is the one-dimensional projection of f (x, y) on the angle θ, i.e., the peak, and s represents the distance of the point from the origin of coordinates. The peaks g (s, theta) correspond to straight lines in the original label image, and the column coordinates theta of these peaks in the Radon transform matrix are the tilt angles of the straight lines perpendicular to the straight lines in the original label image, so the tilt angles of the straight lines in the image are 90-theta.
As a preferred scheme, after the processing unit classifies the tail fibers of the silk ingots, the processing unit performs quality statistics on all the silk ingots on the line to obtain daily inspection total quantity statistics, daily inspection tail fiber number statistics, daily inspection defect rejection quantity statistics and batch abnormal silk statistical information, writes the corresponding information and the current silk ingot detection information into the electronic tag together, and sends the electronic tag to the sorting unit. The electronic tag corresponds to the wire ingot, information of the wire ingot exists in the electronic tag, and the information is sent to the sorting unit, so that the sorting unit can judge the quality of the wire ingot reaching the position of the sorting unit and sort the wire ingot. According to the scheme, the statistical information is written into the electronic tag, and a detailed detection result report is given, so that an operator can better trace the source of the defect according to the statistical information, and the production management and the operation flow are standardized.
As a preferable scheme, the method further comprises a sorting step, which comprises the following steps:
B1. when the silk ingots enter the sorting unit, the sorting unit judges whether the silk ingots are good or not according to the received corresponding classification information of the silk ingots;
B2. if the classified result of the silk ingots is that the tail fibers exist, the silk ingots are judged to be good, the first stop lever is controlled to descend, the silk ingots enter a good product conveying channel, if the silk ingots are classified to be non-tail fibers, the silk ingots are judged to be defective, the second stop lever is controlled to descend, and the silk ingots enter a defective product conveying channel.
Therefore, the invention has the advantages that:
1. the method comprises the following steps of detecting the tail fibers of the silk ingots by adopting a machine vision and image processing technology, and eliminating inferior-quality silk ingots with the tail fibers by arranging a sorting unit, wherein the whole detection system does not influence the normal production of the silk ingots, and a complete set of coherent online processing flow from image acquisition to inferior-quality elimination of the silk ingots is realized;
2. according to the invention, the image acquisition unit and the processing unit extract the label information through an image processing technology, so that the installation of redundant hardware equipment is avoided, the cost is saved, the identity of the silk ingot is determined by sequentially photographing, and the defect detection of the silk ingot is completed;
3. the movement rule of the transmission belt is utilized, and the defective silk ingots are timely, accurately and quickly removed by installing the sorting unit;
4. the measuring and calculating method adopts a deep learning method, so that the speed is higher and the precision is higher.
Drawings
FIG. 1 is a schematic diagram of one configuration of the system of the present invention;
FIG. 2 is a schematic flow chart of the present invention.
The automatic sorting machine comprises a conveyor belt 1, a conveyor belt 2, a tray 3, a camera bellows 4, an image acquisition unit 5, a sorting unit 6, a first stop lever 7, a second stop lever 8, a good product conveying channel 9, a defective product conveying channel 10 and a processing unit.
Detailed Description
The technical scheme of the invention is further specifically described by the following embodiments and the accompanying drawings.
Example (b):
the system for detecting the appearance defects of the chemical fiber tails comprises a tray 2 for loading the silk ingots, a conveyor belt 3 for conveying the tray, labels arranged on the silk ingots, a dark box 3 arranged on the conveyor belt, a sorting unit 5 arranged on the conveyor belt behind the dark box, an image acquisition unit 4 for acquiring label images and silk ingot images arranged in the dark box, and a processing unit 10 for detecting the defects of the tails by the image acquisition unit;
the processing unit is used for identifying and reading label information from the label graph, screening an image to be detected with a silk ingot paper tube from the silk ingot image, carrying out edge detection on a paper tube area of the image to be detected, extracting a main characteristic description paper tube area, and bringing the processed image to be detected into a tail fiber classifier for classification;
and the sorting unit is used for sorting the good and defective products of the silk ingots according to the classification of the tail fibers of the silk ingots.
The tray is used for loading the silk ingots, the tray comprises a base plate and a supporting rod arranged at the center of the base plate, an elastic gripper is arranged on the supporting rod, the silk ingots are sleeved on the supporting rod, the elastic gripper stabilizes the silk ingots on the supporting rod, and the phenomenon that the silk ingots are conveyed in a shaking mode to affect the conveying and imaging quality of the silk ingots is avoided.
The transmission belt is provided with a tray fastening mechanism for fixing and rotating the silk ingots, and the tray fastening mechanism is controlled to be connected with the processing unit. The fastening mechanism comprises a rotating disc, the rotating disc rotates to be controlled by the processing unit, and the rotating disc is provided with a loading and unloading valve, so that the tray can be fastened on the rotating disc through the loading and unloading valve.
The camera comprises a camera group, the camera group comprises a top camera, two side cameras and two bottom cameras, the camera is a CCD (charge coupled device) camera, each camera is internally provided with a photoelectric sensor, and high-precision images of moving objects in a stable time period are acquired by combining an external trigger scanning mode and controllable exposure time. The light source group is mainly used for stably polishing the silk ingots and comprises a top center vertical light source, a bottom left annular light source and a bottom right annular light source, and the three light sources are respectively fixed by a four-angle-shaped angle locking device to prevent the light sources from shaking, so that image imaging is unstable. A photoelectric sensor capable of acquiring high-precision images is arranged in a camera chip, and images of moving objects in a stable time period are acquired by combining an external trigger scanning mode and controllable exposure time. Adopt angle locking mechanism to install the light source group, prevent that vibrations from influencing the light source angle, come to stabilize to the silk spindle and polish, at the upper end of camera bellows and the suitable position and the direction installation camera of bottom. When the filament ingot reaches the exposure range of the camera, the light reflected by the filament ingot is projected onto the sensor through the camera lens, after the sensor is exposed, the photodiode is excited by light to release charges, electric signals are generated, the electric signals control the current generated by the photodiode through a camera chip by utilizing a control signal circuit in a photosensitive element and are output by a current transmission circuit, the camera chip collects the electric signals generated by primary imaging and uniformly outputs the electric signals to an amplifier, the electric signals after amplification and filtering are sent to an A/D (analog/digital) converter, the A/D converter converts the electric signals into digital signals, and outputting the images to a Digital Signal Processor (DSP), carrying out post-image processing such as color correction and white balance processing on the images by the DSP, encoding the images into image files with specific resolution and image format supported by the DC, and finally saving the image files to a memory.
The rear end of the conveying belt is forked to form a good product conveying channel 8 and a defective product conveying channel 9, the sorting unit is arranged at the forked position of the conveying belt and comprises a chassis, a first stop lever 6 located in front of the good product conveying channel and a second stop lever 7 located in front of the defective product conveying channel are respectively arranged on the chassis, the first stop lever and the second stop lever are respectively connected onto an air cylinder, and the air cylinder is controlled to be connected to the processing unit 10. The sorting unit can be controlled by the processing unit to operate, when the filament ingots are detected to be good products, the first stop lever is controlled to descend, the filament ingots are conveyed to the good product conveying channel, when the filament ingots are detected to be defective products, the second stop lever is controlled to descend, and the filament ingots are conveyed to the defective product conveying channel.
A method for detecting appearance defects of chemical fiber tail fibers as shown in figure 2 comprises the following steps,
establishing a fiber classifier for detecting the presence of a fiber, comprising the steps of:
s1, collecting a large number of silk ingot images through an image collecting unit;
s2, carrying out paper tube area positioning and segmentation on the silk ingot image; and preprocessing the image of the silk ingot by positioning and segmenting, wherein the preprocessing is to use median filtering to denoise the image of the silk ingot.
S3, classifying paper tubes in the paper tube area;
s4, carrying out edge detection on the paper tube region image classified into the paper tube, and extracting the main characteristics of the tail fibers;
s5, carrying out tail silk classification on the silk ingot image of the paper tube area processed in the step S4, and sending the well classified silk ingot image of the paper tube area into a bp neural network for training to obtain a tail silk classifier;
and acquiring a label image of the to-be-detected filament ingot, identifying and reading label information, acquiring a filament ingot image of the to-be-detected filament ingot, processing the filament ingot image in steps S2-S4, and classifying the processed filament ingot image through a tail fiber classifier.
According to the embodiment, the machine vision and image processing technology is adopted to detect the tail fibers of the silk ingots in the conveying process of the silk ingots, the sorting unit is arranged to remove defective silk ingots with the tail fibers, the whole detection system does not influence the normal production of the silk ingots, and a set of coherent online processing flow from image acquisition to defective elimination of the silk ingots is realized.
The specific process of positioning and dividing the paper tube region of the silk ingot image in the step S2 includes:
s21, collecting a large number of silk ingot images, carrying out median filtering on the silk ingot images to remove noise, labeling a paper tube area by using a BBox-Label-Tool to generate an external rectangular frame of the paper tube area, taking the labeled silk ingot images as a training set, and converting the training set into data in an LMDB format;
s22, using the pre-trained VGG-16 network provided in the SSD demo as a basic network of the SSD detection network, and carrying in the data set obtained in the step S21 for training; this step uses SSD training to detect paper tube areas. The VGG-16 network is a deep learning network, and after sample training, the silk ingot image of the standard paper tube area is used as a sample, so that the paper tube area of the silk ingot image can be positioned.
And S23, detecting a paper tube area of the silk ingot image by using the SSD, and externally cutting the paper tube area according to the detected rectangle.
The specific process of classifying the paper tubes in the paper tube area in step S3 includes:
s31, collecting a large number of silk ingot images, and taking the paper tube area images of the silk ingot images obtained in the step S2 as a training set;
s32, classifying the paper tube region images into paper tube region images of paper tubes and paper tube region images of non-paper tubes, and training the classified paper tube region images by using a VGG-16 network to obtain a paper tube classifier; the paper tube classifier is a two-classifier, the class 0 represents a paper tube, the class 1 represents a non-paper tube, in the detection process, a loss function layer is removed, an output result of the last layer of the network is extracted, and then a sigmod function is used for normalizing the detection score to be 0-1 to serve as a final detection score.
And S33, sending the paper tube region image to be detected into a paper tube classifier, if the paper tube region image is judged to be a paper tube, entering the step S4, and if the paper tube region image is judged to be a non-paper tube, returning to the step S2, and operating the next silk ingot image. All images of the non-paper tubes are removed in the step.
In step S4, the edge detection is performed on the paper tube region image classified as a paper tube, and the specific process of extracting the main features of the tail fibers includes:
s41, Gaussian filtering is carried out;
s42, calculating the rejection intensity and direction of each pixel point in the paper tube area image;
s43, applying non-maximum value suppression to eliminate the stray corresponding to the edge detection;
s44, determining real and potential edges by applying double-threshold detection;
s45, canny edge detection is completed by inhibiting the encouraged weak edges; in the step, canny edge detection is adopted to highlight the fiber characteristics.
S46, performing principal component analysis on all the paper tube area images, wherein the principal component analysis comprises calculating a covariance matrix of the features, solving the feature values and feature vectors of the covariance matrix, judging the importance degree of the features according to the feature values, selecting a certain number of feature values and feature vectors as the expression of the images, and reducing the extracted feature dimensions; the calculation formula of the assistant is as follows:
C＝∑(x-u)(x-u)T
where C is the covariance matrix, x is the pixel value of the image, and u is the mean of the image. In the step, PCA is adopted to reduce the dimension of the features, and a small number of features are used to describe the paper tube area image, namely the sample, and the features are the features for highlighting the fibers. The main characteristics are extracted, and the classification speed can be accelerated.
The method comprises the following steps of obtaining an image of a to-be-detected filament ingot and simultaneously obtaining a label image of the to-be-detected filament ingot, identifying and reading label information, wherein the specific process comprises the following steps:
A1. setting each rotation angle of the silk ingot through calculation, and acquiring a label image by an image acquisition unit after the silk ingot rotates by one angle; the specific process for acquiring the image with the label comprises the following steps:
A11. setting each rotation angle of the silk ingot through calculation, rotating the silk ingot when the silk ingot enters an image acquisition area, and acquiring a label image by an image acquisition unit after the silk ingot rotates for an angle;
A12. performing binarization and corrosion expansion processing on a label image to obtain an image connected region, calculating the area size of the connected region, judging whether a label exists according to a preset prior threshold value [ AreaMin, AreaMax ] according to whether the area of the connected region falls within the range of the prior threshold value [ AreaMin, AreaMax ], and if the area of the connected region does not fall within the prior range, judging that the label does not exist, continuing rotating the silk spindle until the label is judged to exist; and if the label exists, calculating the offset angle of the center position of the connected region, and rotating the label to the center region of the camera image. Placing the tag in the center of the camera prevents loss of accuracy due to angular misalignment.
A2. Pre-building segmentation models
Collecting a large number of label images as a training set, labeling the shape of a label area and an external rectangular frame on the label images, and carrying out segmentation training by taking data into the FCN; FCN (full volumetric Networks for semantic segmentation) is a network structure based on full convolution to realize target segmentation, and adopts an end-to-end segmentation structure, inputs a whole graph and outputs a segmentation result, and the segmentation result expresses a target area in the form of an area set.
A3. Sending the label image to be tested into an FCN test, wherein the test result of the FCN is the position of the label, and obtaining a segmented label area image; and finding out the structure which best accords with the practical application scene as the position of the label according to the ROC curve. The receiver operating characteristic curve is abbreviated as ROC curve.
And correcting the obtained segmented label area image, wherein the steps comprise:
A31. cutting a label area image from an original label image according to the FCN segmentation result to be used as an interested area, binarizing the interested area image by using an edge function, and detecting the boundary position of the label area by using the binarized image;
A32. using a Radon corrected inclined character label, calculating Radon transformation of the edge image, and performing Radon transformation on each pixel of the position of the inclined character in the label area image, wherein the pixel is 1; the Radon algorithm is an algorithm for determining an image inclination angle by finding an angle at the maximum projection value through directional projection superposition. The projections are respectively made in the directions of 0-179 degrees. The training is facilitated by correcting the size of the image of the same label area.
A33. Detecting a peak value of g (s, theta) in the Radon transformation matrix, correcting the label area according to the peak value, wherein the relation between a Radon transformation plane and an element plane is as follows:
where f (x, y) is the gray scale value of the point (x, y) on the original image plane, g (s, θ) is the one-dimensional projection of f (x, y) on the angle θ, i.e., the peak, and s represents the distance of the point from the origin of coordinates. The peaks g (s, theta) correspond to straight lines in the original label image, and the column coordinates theta of these peaks in the Radon transform matrix are the tilt angles of the straight lines perpendicular to the straight lines in the original label image, so the tilt angles of the straight lines in the image are 90-theta.
A4. Pre-building recognition models
Collecting a large number of label images to perform step A3 to obtain a label area image, extracting the characteristics of the label area image by using a VGG-16 network, copying N parts of the processing result of the label area image, and sending the copied result into a recurrent neural network for recognizing the symbols with indefinite length to train;
A5. and sending the image of the area of the label to be detected into a recurrent neural network for label character recognition, and acquiring label information.
The processing unit writes the tail yarn detection information into a newly-built electronic tag, performs quality statistics on all the wire ingots on the wire after classifying the wire ingots, obtains daily inspection wire ingot total amount statistics, daily inspection tail yarn number statistics, daily inspection defect rejection amount statistics and batch abnormal yarn statistical information, writes the corresponding information into the electronic tag and stores the information, and sends the electronic tag to the sorting unit.
And when the silk ingots reach the position of the sorting unit, the sorting unit sorts the silk ingots according to the received electronic tag information corresponding to the silk ingots. The specific processing unit can calculate the time of the current spindle reaching the sorting unit according to the conveying speed of the conveying belt and the distance from the current position of the spindle to the sorting unit, and the time is written into the electronic tag, so that the sorting unit can obtain the information of the corresponding spindle when the spindle reaches the sorting unit.
The sorting step comprises:
B1. when the silk ingots enter the sorting unit, the sorting unit judges whether the silk ingots are good or not according to the received corresponding classification information of the silk ingots;
B2. if the classified result of the silk ingots is that the tail fibers exist, the silk ingots are judged to be good, the first stop lever is controlled to descend, the silk ingots enter a good product conveying channel, if the silk ingots are classified to be non-tail fibers, the silk ingots are judged to be defective, the second stop lever is controlled to descend, and the silk ingots enter a defective product conveying channel.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.
Although the terms conveyor belt, tray, camera bellows, image acquisition unit, sorting unit, etc. are used more often herein, the possibility of using other terms is not excluded. These terms are used merely to more conveniently describe and explain the nature of the present invention; they are to be construed as being without limitation to any additional limitations that may be imposed by the spirit of the present invention.
Claims (9)
1. A chemical fiber filament tail appearance defect detection system is characterized in that: the automatic tail fiber defect detection device comprises a tray for loading a silk ingot, a conveyor belt for conveying the tray, a label arranged on the silk ingot, a camera bellows arranged on the conveyor belt, a sorting unit arranged on the conveyor belt behind the camera bellows, an image acquisition unit for acquiring a label image and a silk ingot image arranged in the camera bellows, and a processing unit for carrying out tail fiber defect detection by sending acquired information to the image acquisition unit;
a processing unit for identifying and reading the label information from the label image, screening out the image to be measured with the silk spindle paper tube from the silk spindle image,
carrying out edge detection on a paper tube area of an image to be detected, extracting main characteristic description paper tube areas, and the specific process comprises the following steps:
carrying out Gaussian filtering;
calculating the rejection intensity and direction of each pixel point in the paper tube area image;
applying non-maximum suppression to eliminate the spurious response coming from edge detection;
determining true and potential edges using dual threshold detection;
canny edge detection is done by suppressing encouraged weak edges; in the step, canny edge detection is adopted to obviously show the characteristics of the fibers,
performing principal component analysis on all the images of the paper tube area, wherein the principal component analysis comprises calculating a covariance matrix of the features, solving characteristic values and characteristic vectors of the covariance matrix, judging the importance degree of the features according to the size of the characteristic values, selecting a certain number of characteristic values and characteristic vectors as the expression of the images, and reducing the extracted characteristic dimensions; the calculation formula of the covariance is as follows:
C＝∑(x-u)(x-u)T
wherein C is a covariance matrix, x is a pixel value of the image, and u is a mean value of the image;
the processed image to be detected is taken into a tail fiber classifier for classification;
and the sorting unit is used for sorting the good and defective products of the silk ingots according to the classification of the tail fibers of the silk ingots.
2. The system as claimed in claim 1, wherein the rear end of the conveyor belt is branched to form a good product conveying passage and a bad product conveying passage, the sorting unit is disposed at the branched portion of the conveyor belt, the sorting unit comprises a base plate, a first stop lever and a second stop lever are respectively disposed on the base plate and located in front of the good product conveying passage and the bad product conveying passage, the first stop lever and the second stop lever are respectively connected to a cylinder, and the cylinder is controlled to be connected to the processing unit.
3. A method for detecting cosmetic defects in a chemical fiber tail using the system of claim 1 or 2, wherein: comprises the steps of (a) preparing a mixture of a plurality of raw materials,
establishing a fiber classifier for detecting the presence of a fiber, comprising the steps of:
s1, collecting a large number of silk ingot images through an image collecting unit;
s2, carrying out paper tube area positioning and segmentation on the silk ingot image;
s3, classifying paper tubes in the paper tube area;
s4, carrying out edge detection on the paper tube region image classified into the paper tube, and extracting the main characteristics of the tail fibers; the specific process comprises the following steps:
s41, Gaussian filtering is carried out;
s42, calculating the rejection intensity and direction of each pixel point in the paper tube area image;
s43, applying non-maximum value suppression to eliminate the stray corresponding to the edge detection;
s44, determining real and potential edges by applying double-threshold detection;
s45, canny edge detection is completed by inhibiting the encouraged weak edges;
s46, performing principal component analysis on all the paper tube area images, wherein the principal component analysis comprises calculating a covariance matrix of the features, solving the feature values and feature vectors of the covariance matrix, judging the importance degree of the features according to the feature values, selecting a certain number of feature values and feature vectors as the expression of the images, and reducing the extracted feature dimensions; the calculation formula of the covariance is as follows:
C＝∑(x-u)(x-u)T
wherein C is a covariance matrix, x is a pixel value of the image, and u is a mean value of the image;
s5, carrying out tail yarn classification on the silk ingot image of the paper tube area processed in the step S4, and sending the well classified silk ingot image of the paper tube area into a bp neural network training tail yarn classifier;
and acquiring a label image of the to-be-detected filament ingot, identifying and reading label information, acquiring a filament ingot image of the to-be-detected filament ingot, processing the filament ingot image in steps S2-S4, and classifying the processed filament ingot image through a tail fiber classifier.
4. The method as claimed in claim 3, wherein the specific process of positioning and dividing the paper tube region of the filament beam image in step S2 comprises:
s21, collecting a large number of silk ingot images, labeling a paper tube area by using a BBox-Label-Tool, generating an external rectangular frame of the paper tube area, taking the labeled silk ingot images as a training set, and converting the training set into data in an LMDB format;
s22, using the pre-trained VGG-16 network provided in the SSD demo as a basic network of the SSD detection network, and carrying in the data set obtained in the step S21 for training;
and S23, detecting a paper tube area of the silk ingot image by using the SSD, and cutting out the paper tube area according to the detected rectangular outside.
5. The method as claimed in claim 4, wherein the step of classifying the paper tube in the paper tube area in step S3 comprises:
s31, collecting a large number of silk ingot images, and taking the paper tube area images of the silk ingot images obtained in the step S2 as a training set;
s32, classifying the paper tube region images into paper tube region images of paper tubes and paper tube region images of non-paper tubes, and training the classified paper tube region images by using a VGG-16 network to obtain a paper tube classifier;
and S33, sending the paper tube region image to be detected into a paper tube classifier, if the paper tube region image is judged to be a paper tube, entering the step S4, and if the paper tube region image is judged to be a non-paper tube, returning to the step S2, and operating the next silk ingot image.
6. The method for detecting the appearance defects of the chemical fiber tail fibers as claimed in claim 3, wherein the specific processes of obtaining the label image of the to-be-detected fiber spindle and identifying and reading the label information comprise:
A1. setting each rotation angle of the silk ingot through calculation, and acquiring a label image by an image acquisition unit after the silk ingot rotates by one angle;
A2. pre-building segmentation models
Collecting a large number of label images as a training set, labeling the shapes of label areas and external rectangular frames on the label images, and carrying out segmentation training by bringing data into the FCN;
A3. sending the label image to be tested into an FCN test, wherein the test result of the FCN is the position of the label, and obtaining a segmented label area image;
A4. pre-building recognition models
Collecting a large number of label images to perform step A3 to obtain a label area image, extracting the characteristics of the label area image by using a VGG-16 network, copying N parts of the processing result of the label area image, and sending the copied result into a recurrent neural network for recognizing the symbols with indefinite length to train;
A5. and sending the image of the area of the label to be detected into a recurrent neural network for label character recognition, and acquiring label information.
7. The method as claimed in claim 6, wherein the step of obtaining the divided label region image and then correcting the image in step A3 comprises:
A31. cutting a label area image from an original label image according to the FCN segmentation result to be used as an interested area, binarizing the interested area image by using an edge function, and detecting the boundary position of the label area by using the binarized image;
A32. using a Radon corrected inclined character label, calculating Radon transformation of the edge image, and performing Radon transformation on each pixel of the position of the inclined character in the label area image, wherein the pixel is 1;
A33. detecting a peak value of g (s, theta) in the Radon transformation matrix, correcting the label area according to the peak value, wherein the relation between a Radon transformation plane and an element plane is as follows:
where f (x, y) is the gray scale value of the point (x, y) on the original image plane, g (s, θ) is the one-dimensional projection of f (x, y) on the angle θ, i.e., the peak, and s represents the distance of the point from the origin of coordinates.
8. The method as claimed in claim 3, wherein the processing unit classifies the tails of the filament ingots, performs quality statistics on all the filament ingots on the line to obtain daily inspection total number statistics, daily inspection defect rejection statistics and batch abnormal filament statistical information, writes the corresponding information and the current detection information of the filament ingots into the electronic tag and stores the information, and sends the electronic tag to the sorting unit.
9. The method as claimed in claim 3, further comprising a sorting step comprising:
B1. when the silk ingots enter the sorting unit, the sorting unit judges whether the silk ingots are good or not according to the received corresponding classification information of the silk ingots;
B2. if the classified result of the silk ingots is that the tail fibers exist, the silk ingots are judged to be good, the first stop lever is controlled to descend, the silk ingots enter a good product conveying channel, if the silk ingots are classified to be non-tail fibers, the silk ingots are judged to be defective, the second stop lever is controlled to descend, and the silk ingots enter a defective product conveying channel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811635403.8A CN109550712B (en) | 2018-12-29 | 2018-12-29 | Chemical fiber filament tail fiber appearance defect detection system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811635403.8A CN109550712B (en) | 2018-12-29 | 2018-12-29 | Chemical fiber filament tail fiber appearance defect detection system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109550712A CN109550712A (en) | 2019-04-02 |
| CN109550712B true CN109550712B (en) | 2020-09-22 |
Family
ID=65871897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811635403.8A Active CN109550712B (en) | 2018-12-29 | 2018-12-29 | Chemical fiber filament tail fiber appearance defect detection system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109550712B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689520A (en) * | 2019-08-26 | 2020-01-14 | 浙江华是科技股份有限公司 | Magnetic core product defect detection system and method based on AI |
| CN110647851B (en) * | 2019-09-27 | 2022-12-09 | 普联技术有限公司 | Production line capacity monitoring method, device and system |
| CN111398287A (en) * | 2019-11-29 | 2020-07-10 | 合肥国轩高科动力能源有限公司 | Battery pole piece scratch detection system and detection method |
| CN111141741B (en) * | 2019-12-09 | 2023-02-10 | 合肥中科迪宏自动化有限公司 | Detection apparatus capable of identifying label defects of printer |
| CN110956656A (en) * | 2019-12-17 | 2020-04-03 | 北京工业大学 | Spindle positioning method based on depth target detection |
| CN111415343B (en) * | 2020-03-18 | 2020-10-27 | 珠海市奥德维科技有限公司 | Artificial intelligence-based six-side appearance detection method for chip multilayer ceramic capacitor |
| CN113533358B (en) * | 2020-04-16 | 2023-12-22 | 杭州慧知连科技有限公司 | Visual detection method and device for surface shape defects of silk ingot |
| CN113533357B (en) * | 2020-04-16 | 2023-09-29 | 杭州慧知连科技有限公司 | Visual inspection method and inspection device for surface defects of silk ingot |
| CN113533359A (en) * | 2020-04-16 | 2021-10-22 | 杭州慧知连科技有限公司 | Visual detection method and device for surface defects of silk ingots |
| CN111583202B (en) * | 2020-04-27 | 2023-09-01 | 浙江华睿科技股份有限公司 | Method and device for detecting broken filaments |
| CN111815564B (en) * | 2020-06-09 | 2024-03-29 | 浙江华睿科技股份有限公司 | Method and device for detecting silk ingots and silk ingot sorting system |
| CN112084964A (en) * | 2020-09-11 | 2020-12-15 | 浙江水晶光电科技股份有限公司 | Product identification apparatus, method and storage medium |
| CN112330595B (en) * | 2020-10-13 | 2024-04-02 | 浙江华睿科技股份有限公司 | Method and device for detecting stumbling wire, electronic equipment and storage medium |
| CN112505049B (en) * | 2020-10-14 | 2021-08-03 | 上海互觉科技有限公司 | Mask inhibition-based method and system for detecting surface defects of precision components |
| CN112396344A (en) * | 2020-11-30 | 2021-02-23 | 天津大学 | Chemical process reliability online evaluation method based on product quality |
| CN113588686A (en) * | 2021-02-10 | 2021-11-02 | 菲特(珠海横琴)智能科技有限公司 | Optical design system based on silk ingot visual detection and polishing and light and dark field correction method |
| CN113189103B (en) * | 2021-04-30 | 2024-04-30 | 杭州辰泽新材料有限公司 | Online detection system and fault diagnosis method for Y fine denier cationic polyester filaments |
| CN113129305B (en) * | 2021-05-18 | 2023-06-16 | 浙江大华技术股份有限公司 | Method and device for determining state of silk spindle, storage medium and electronic device |
| CN113560199A (en) * | 2021-06-25 | 2021-10-29 | 普洛斯企业发展(上海)有限公司 | Sorting method of logistics tray |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH11258170A (en) * | 1998-03-13 | 1999-09-24 | Techno Soft Systemnics:Kk | Automatic inspection apparatus for residue and abrasion at inside of mouthpiece for chemical fiber inspection |
| US7952719B2 (en) * | 2007-06-08 | 2011-05-31 | Prescient Medical, Inc. | Optical catheter configurations combining raman spectroscopy with optical fiber-based low coherence reflectometry |
| JP6019798B2 (en) * | 2012-06-22 | 2016-11-02 | ソニー株式会社 | Information processing apparatus, information processing system, and information processing method |
| CN104050450A (en) * | 2014-06-16 | 2014-09-17 | 西安通瑞新材料开发有限公司 | Vehicle license plate recognition method based on video |
| CN104792789A (en) * | 2015-04-08 | 2015-07-22 | 上海常良智能科技有限公司 | Chemical fiber paper tube appearance detection device and method |
| CN106022805B (en) * | 2016-05-25 | 2019-09-24 | 华中科技大学 | A kind of anti-fake traceability system and method based on tag readable degree |
| CN106841209B (en) * | 2016-12-28 | 2019-05-24 | 杭州慧芯智识科技有限公司 | One kind being based on big data self study chemical fiber wire ingot intelligence appearance detection system and method |
| CN108508035B (en) * | 2018-05-10 | 2023-11-24 | 杭州慧知连科技有限公司 | Intelligent detection equipment for appearance defects of chemical fiber silk ingots |
| CN108985294B (en) * | 2018-07-13 | 2022-03-25 | 广东工业大学 | Method, device and equipment for positioning tire mold picture and storage medium |
-
2018
- 2018-12-29 CN CN201811635403.8A patent/CN109550712B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109550712A (en) | 2019-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109550712B (en) | Chemical fiber filament tail fiber appearance defect detection system and method | |
| CN109454006B (en) | Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects | |
| CN109255787B (en) | System and method for detecting scratch of silk ingot based on deep learning and image processing technology | |
| CN115351598A (en) | Numerical control machine tool bearing detection method | |
| CN110211101A (en) | A kind of rail surface defect rapid detection system and method | |
| CN117635565B (en) | Semiconductor surface defect detection system based on image recognition | |
| CN113608378B (en) | Full-automatic defect detection method and system based on LCD (liquid crystal display) process | |
| CN101063659A (en) | System for detecting paper flaw of paper sheet | |
| CN110473184A (en) | A kind of pcb board defect inspection method | |
| CN104483320A (en) | Digitized defect detection device and detection method of industrial denitration catalyst | |
| CN113177924A (en) | Industrial production line product flaw detection method | |
| CN113222938A (en) | Chip defect detection method and system and computer readable storage medium | |
| CN113324864A (en) | Pantograph carbon slide plate abrasion detection method based on deep learning target detection | |
| CN111398287A (en) | Battery pole piece scratch detection system and detection method | |
| CN113820319A (en) | Textile surface defect detection device and method | |
| CN111426693A (en) | Quality defect detection system and detection method thereof | |
| CN114662594B (en) | Target feature recognition analysis system | |
| CN116434206A (en) | Cotton quality character detection method based on machine vision technology | |
| CN113228049B (en) | Milk analyzer for classifying milk | |
| CN109622404B (en) | Automatic sorting system and method for micro-workpieces based on machine vision | |
| CN113793322A (en) | Method for automatically detecting magnetic material, electronic equipment and storage medium | |
| CN106501278B (en) | Surface of the light tube defect classification method and system based on invariable rotary textural characteristics | |
| CN114596242A (en) | Defect detection method, defect detection device, electronic equipment and computer readable storage medium | |
| CN111398138A (en) | Optical detection system and method of dry type blood cell analysis device | |
| Zhang et al. | Defect detection of mobile phone screen based on improved difference image method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |